Chapter 06: Dr. Gehan's Research and Publishing Impact
Files
Loading...
Description
In this chapter, Dr. Gehan assesses his research output. He talks about: the number of patients required in Phase II clinical trial, comparison of two survival distributions with no average survival time, the extension of the Wilcoxin (Frank) Test with two-sample tests with censoring (loss) when working with Professor Sir David Roxbee Cox in London, the Gehan-Breslow (Norman Edward)-Wilcoxon generalization of the Wilcoxon statistic test to the two-sample problem with censored data, and hazard functions. Dr. Gehan then discusses the impact of computers on statistical research publications. He talks about the necessity for, and research perspectives and publications regarding, non-randomized controls in cancer clinical trials. He mentions the influence of Dr. Franz Joseph Ingelfinger, Editor of the New England Journal of Medicine, in promoting a constructive Gehan/Freireich publication approach where there was a conflicting research theory (Thomas C. Chalmers, known for randomized clinical trials). He also discusses the development of cancer research cooperative groups, epidemiologist Eleanor Macdonald and many of his co-authored biostatics articles.
Identifier
GehanE_01_20030328_C06
Publication Date
2003
Publisher
The Making Cancer History® Voices Oral History Collection, The University of Texas MD Anderson Cancer Center
City
Houston, Texas
Interview Session
Edmund A. Gehan, PhD, Oral History Interview, March 28, 2003
Topics Covered
The Interview Subject's Story - The Researcher; The Researcher; Research; Human Stories; Funny Stories; Collaborations; Giving Recognition; Understanding Cancer, the History of Science, Cancer Research
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License.
Disciplines
History of Science, Technology, and Medicine | Oncology | Oral History
Transcript
Edmund A. Gehan, PhD:
I will just try to give you a little bit of the background. This reference one (Reference #1?) is how many patients to you need in a Phase II trial? There was a big search at NCI before coming to M. D. Anderson, you know. There was a whole pipeline, studies in animals, pharmacology, and so on. I would be at meetings like this: the physician would study five patients; they were all toxic; the treatment is worthless, let’s stop. So I addressed the question of what is the minimum number of patients you should study, that if they all fail, then you can stop. One answer to that question is fourteen, which is justified in the paper. Fourteen has become almost a magic number in Phase II trials. As I say, I have a copy of this paper. Reference two (Reference #2) is a test for survival data, and it arose directly from the clinical trial that I did with Frei and Freireich. This is perhaps the most theoretical paper that I have written. Well, then I guess in a way it is some justification of “why do you need a statistician working in a clinical trial?” It’s a way to compare two survival distributions. A lot of the patients are still alive in both groups, so there is no average survival time. I worked on this paper in London, working with Professor Cox, and both of these are citation classics. Somehow if you are going to have a test named after you, you’d have it made. (Laughter)
Lesley W. Brunet:
Yes.
Edmund A. Gehan, PhD:
These are some pages from the Encyclopedia of Biostatistics [References 1 and 2]. So there were games designed for Phase II trials. I called it a generalization of Wilcoxin test, but since then people have called it the Gehan test. My proposals were basically the first ones in both of these fields. People have gone on further with a lot, obviously a lot further. And that is spelled out in the write-up that is in the Encyclopedia [of Biostatistics]. Well, I am proud of those parts. Believe it or not, in statistics, there is a book called Data. What they said is, “It’s not always easy to find a good example that you can work with.” Agnes Hertzberg is a good friend of mine. Her own handicap is that her father won the Nobel Prize and she will never live up to her father, but she is a very fine woman. Anyway, she asked me to use this leukemia data, and actually this is a very readable thing, but it shows how the data that got into that Wilcoxin test. This not only got into the Wilcoxin test, but I said I worked with David Cox in London. The most famous paper published in the twentieth century was Cox’s paper on regression models and life tables. In that paper he uses this data as an example. By the way, if I was smart enough, I would have developed that test, but I didn’t. (Laughter) Anyway, this is a readable thing as to how this clinical trial of cancer lead to my work on comparing survival distributions, and also to Cox’ work. If he is asked, “Where did you get the idea for that?” He doesn’t say, “I got it from Gehan.” He doesn’t say that. There were several other influences, too. And as I said, he was brilliant. This is a paper that I wrote at M. D. Anderson [Reference 3]. People don’t do things like this anymore. An important aspect of survival functions is estimating the hazard function, the risk of death per unit of time. Actually Cox’s work is based upon dealing with hazard function. If you group all of the survival into intervals, it tells you how to plot the hazard function. Here is the hazard function. You can see the risk was high here, then it was flattened out then it went up. I think this was a nice paper. But people don’t do this anymore because [with] computers, you don’t have to group the data. This paper, “Non-randomized Controls in Cancer Clinical Trials” (Reference #) is a special article in the New England Journal [of Medicine]. This study is one of the most quoted sets of data. This study was the first prospective, randomized, double blind, placebo-controlled, sequential analysis theory. It was the first study of this type to be done and Freireich was the co-author. His was the main idea for the clinical study. Freireich’s proposal since then said, “Well, I’ve been there, done that and now I don’t need to do that anymore.” All the combination therapy, they didn’t randomize, they didn’t randomize. I guess you’d have to ask him something about randomization; there were certain things on the skid about that.
Lesley W. Brunet:
Yes.
Edmund A. Gehan, PhD:
There was a paper published before this by Tom Chalmers, “Controls in Clinical Investigations” in which he said, “There was a good argument for randomizing the very first patient in a new therapy. And if the study wasn’t randomized, it wasn’t controlled.” Well, I didn’t believe that. Dr. Ingelfinger was then the Editor of the New England Journal [of Medicine]. We said we wanted to comment on Chalmers’ paper, and he said to his credit, “I don’t want anything saying that Chalmers is full of bologna. Give me a positive paper of why non-randomized controls are good.” So we did. I basically wrote the paper. This was a key justification for why you might want to consider non-randomizing. I know you have been treated for cancer, but not many people want to have their treatment decided by the flip of a coin. Although I statistically support randomized trials when there is equipoise. One of the stories is that Dr. …I’ll have to think of his name again now. He was in pediatrics and he had to deal with parents that had a child with cancer.
Lesley W. Brunet:
Was he here?
Edmund A. Gehan, PhD:
No, he wasn’t [here]. He was in Philadelphia. He was trying to persuade the parents to enter their child on this randomized trial. The parents said, “Which one do you think is better, doctor?” He said, “I don’t know which one is better, that’s why we want to do the randomized trial. We’ll find out which one is better.” “Yes, doctor, I know that you don’t know, but which one do you think is better?” And with that he had to say, “Well, this one may be better.” “Well, that’s the one I want for my kid.” Logically there are many reasons for randomized trials and theoretically I am for them. Freireich I think goes too far the other way, that you never have to randomize. This is another rational basis for historical controls, the logic for doing it. These were all written when I was here. (Ref ?) There is one statistical evaluation of therapies and historical controls (Ref ?). One of the points this makes is, “Let’s look at the progress in cancer.” There is a table here. What new treatments were found by randomized versus non-randomized trials? Here is a list that came out of historical control studies. Some of them weren’t well historically controlled, and one of them was by Watsuto in osteosarcoma. This may not be a great paper, but “The Training of Statisticians for Cooperative Clinical Trials” I did write while I was here (Ref ?). As you get older, how do you teach somebody coming into the field? This is a series of thoughts about that, some of which came from Mantel. It said, “Beware of people that say I only need five minutes of your time to handle my [problem].” They’ve been thinking about it for years maybe, and they only need five minutes of your time to solve their problem. Beware of such people; they need more than five minutes. This is an article also written here, “A Strategy for Evaluation of New Treatments in Untreated Patients: Application to a Clinical Trial of AMSA for Acute Leukemia” [Reference 10]. It was Freireich’s basic idea. We all want to evaluate new treatments, but there are already treatments available for all kinds of cancer. So if a new patient comes in, and those are the ones that you really want most, how do you try to evaluate a new treatment in that patient? The standard way of doing it is administer all the known things and then when things are toward the end of the line, then give the new [treatment]. Well, that is the biggest hurdle for the new treatment. Maybe you’d be better giving it closer up front. This was an idea that Freireich had, but we implemented, which said, “Use a statistical regression model based upon—in this case—their age, their temperature, antecedent hematologic disorder, BUN, hemoglobin, liver size.” We can use this model based on past data and get out of it a probability that you are going to respond to the treatment, given the standard. Some of the patients are going to have a high probability, and those you give the standard. Those that have a low probability, you give them the new [treatment]. Tape 2, Side B
Edmund A. Gehan, PhD:
As I say, administering the new treatment to those with lower probability of response. Actually, it turned out they did pretty well, and the criteria were gradually moved up. I think it’s a good idea that hasn’t adequately been followed. This is “Historical and Methodological Developments in Clinical Trials at the National Cancer Institute” (Reference ?). This paper was written here in 1989, and this is somewhat of an historical paper. How do the cooperative groups develop? You take the cooperative groups program in 1960, and at that time [in] the Southwest Group, Dr. Grant Taylor was the chairman and Eleanor Macdonald was the statistician. This takes all the groups. Frei was acute leukemia. He was the chair of that one and Schneiderman was the statistician. I am not listed on any of these, but a year later I had taken Schneiderman’s place. This gives a pretty good historical review of the clinical trials program as it evolved at NCI and spread out. James S. Olson, PhD Was Eleanor Macdonald a statistician?
Edmund A. Gehan, PhD:
No. I guess she has died now.
Lesley W. Brunet:
No, she is living. James S. Olson, PhD Yes, still alive.
Edmund A. Gehan, PhD:
Good for her. I think she would consider herself an epidemiologist. She was one of the old buddies of Dr. Clark, and she worked for the Connecticut Tumor Registry. She wasn’t trained in mathematics or statistics. I think she certainly knew some statistics, but a lot of hospitals have tumor registries, and that was the background from which she came. James S. Olson, PhD Right.
Edmund A. Gehan, PhD:
I think she had passed over the statistical part of the group to Ken Griffith, even by the time that I came here. We were always pretty good friends although we didn’t work very closely together. If you didn’t have any proprietary interest in maintaining the head of the statistical (trails off). James S. Olson, PhD Who replaced her in epidemiology?
Lesley W. Brunet:
(?) (Counter 48)
Edmund A. Gehan, PhD:
He came well after she was out of the picture, I think. I think there was an interval. You’d have to ask Genet [Louis]. I am sure he is still around. He is retired. Who replaced her [Eleanor Macdonald]? Genet eventually did, but I think there was an interval there. I don’t know what it was.
Lesley W. Brunet:
Do you need to run?
Edmund A. Gehan, PhD:
No. Well, I guess in five minutes. This is the article I wrote for that encyclopedia on nonrandomized trials. It gives arguments against and for, “Under what conditions might you really do it?” As I said, a couple of years ago they had “The Role of the Biostatistician in Cancer Research.” I gave this talk at M. D. Anderson [2000] about “Triumphs and Challenges in Oncology.” This was a review statistics and the development of statistics in one slide. This was abstracted from another publication (Ref?). The key people at NIH, and this is a whole story by itself. This is a group of consultants, Sam Greenhouse, who died two years ago, Nathan Mantel, who died last summer, Jerry [Jerome] Cornfield, who was treated the last few months of his life here at M. D. Anderson, Max Halperin, Marvin Schneiderman, and Harold Dorn. Harold Dorn was the head of this group. Greenhouse was from New York, Mantel, Leiberman, Cornfield, Marvin Schneiderman; Max Halperin was from Omaha, Nebraska. Anyway, they were in this room and somebody would call up and one of them would take on whatever the project was. This sort of gives some of the ideas of their philosophy. Jerry Cornfield was the first one to support the ideas of relative risk. In the lung cancer/smoking studies, what’s easy to do is to look at a group of people with lung cancer, another group without lung cancer, and ask them whether they smoke. But that is not the question you want answered. The question you want answered is, “If you smoke, what is your risk of getting lung cancer?” He developed and justified the first relative risk estimate. There is a book on Breakthroughs in Biostatistics that I refer to here, and he was a great guy who died of pancreatic cancer here in 1979. And that is almost another story because he was treated here by J [Emil J] Freireich three of the last four weeks of his life. The other one was Mantel, who I learned a lot from, and as I say he was my first boss. There is a Mantel and Hensell (?) paper that is a citation classic that he wrote in 1959. The third one is David Cox. He was knighted in 1985. People don’t talk about the Cox model now, but where are we going to go from here? This was some effort at saying, “What is going to happen to biostatistics in the new millenium?” I mentioned before that I worked at this resort in New Jersey, but I have always been interested in humorous things, too. (Alarm sounds)
Lesley W. Brunet:
Is that your alarm?
Edmund A. Gehan, PhD:
One of my papers (trails off).
Lesley W. Brunet:
I haven’t seen that one.
Edmund A. Gehan, PhD:
You haven’t seen this? “How to Find the Cure for Cancer”? You want to go directly to the problem, right? How to find the cure for cancer. So I have a paper that is published on that (Ref ?). We [the Southwest Oncology Group] were doing pretty well and we were meeting in Las Vegas. Usually I gave a lecture on some statistical topic. [I thought] I’ll have to do something different in this meeting in Las Vegas, “How to Find the Cure for Cancer.” Well, I can tell you the answer is statistics is the cure for cancer. If you manipulate your statistics properly, you can cure cancer. (Laughter) I am proud of this article. This was a lead article in the Journal of Irreproducible Results.
Lesley W. Brunet:
Was that the name of the journal actually?
Edmund A. Gehan, PhD:
It was a real journal. There is a copy of every one of those reprints.
Lesley W. Brunet:
After your luncheon, do you want to come back?
Edmund A. Gehan, PhD:
Yes, I’d be glad to. I kind of enjoy this, as you might have gathered here. I have done a lot of the talking here.
Lesley W. Brunet:
Why don’t we interrupt this for now—we’ll stop this for now, and we’ll talk about it (? Counter 133). [Session Resumes]
Recommended Citation
Gehan, Edmund A. PhD; Brunet, Lesley W.; and Olson, James S. PhD, "Chapter 06: Dr. Gehan's Research and Publishing Impact" (2003). Interview Chapters. 313.
https://openworks.mdanderson.org/mchv_interviewchapters/313
Conditions Governing Access
Open
